Is an embedded CTO or CAIO better than hiring a strategy firm?

They solve different problems. An outside-in review can be useful when you need a fresh view of the market or a one-off decision. It is a poor fit for AI adoption, where the plan changes every few months as models change, and where the hard part is the building and the change in how people work. An embedded owner stays through the plan, the build and the hand-over, so the person who made the promise is the one who has to keep it.

All of Pulse

Nobody owns the number: why AI plans die in the handoff

Q: Why do AI plans fail in small companies?

Usually not because the plan was wrong or the technology failed, but because nobody owned the result. The plan is written by one party, the tool is sold by another and the integration is done by a third. Each does its part properly, and each hands over to the next. The outcome the business wanted — fewer hours on a task, faster replies, lower cost — belongs to none of them, so nobody is measuring it and nobody changes course when it slips.

Q: What does it mean to own the number?

It means one named person is accountable for the business outcome an AI project was meant to change, not for a deliverable along the way. They agree the number up front, they can see it every week, and they have the authority to change the plan, the tool or the process when it is not moving. Without that person, a project can deliver everything it promised and still change nothing.

ExponenLabs24 September 20267 min read

There is a pattern we see often enough to have a name for it. A company decides it needs an AI plan. It pays for one and gets a deck. Then nobody stays to build it.

The deck is usually good. It names the right opportunities, sizes them sensibly, and has a roadmap on the last slide. Six months later the roadmap has been partly built, by different people, and the number the deck promised to move has not moved. Nobody can quite say why, because nobody was watching it.

This is not a story about bad work. Every party in it did their job. It is a story about a chain with no one holding the end of it.

The three-step chain

For a small or mid-sized company, an AI project usually passes through three sets of hands.

The deck. A strategy firm, or an internal working group, looks at the business from the outside in. It interviews people, reviews processes, benchmarks the market, and writes down where AI could help. Its job ends when the deck is presented.

The vendor. Someone sells a tool that matches a slide. A support bot, a sales assistant, a document platform. The vendor's job is to make the tool work as a product. Whether it changes your business is outside their contract.

The integrator. A development shop or the internal IT team connects the tool to your systems. Their job is to ship the integration to spec. The spec came from the vendor's documentation and the deck's slide, and nobody who wrote either is still in the room.

Each handoff loses something. The deck knows why a workflow matters but not how the tool behaves. The vendor knows the tool but not the workflow. The integrator knows the systems but not the reason. By the end, the thing that was meant to change — hours spent, reply time, error rate, cost — has passed through three owners and belongs to none of them.

Why AI makes the handoff worse

This chain has been around for as long as companies have bought technology. AI makes it more fragile, for three reasons.

The plan goes stale quickly. Models, prices and capabilities change every few months. A plan written in spring can be wrong about what is cheap or possible by autumn. A deck is a snapshot. AI adoption needs someone who updates the plan as the ground moves.

The hard part is after the build. Traditional software mostly works the same way on day 300 as on day one. An AI workflow does not. Its inputs drift, its model gets updated, and it can get quietly worse without any error. Someone has to watch its output, run the checks and fix it. That work starts after the integrator has left.

The change is in how people work. An AI workflow changes who does what. The support team reviews drafts instead of writing replies. Finance approves exceptions instead of reconciling every line. If nobody owns that change — the training, the decision rights, the new definition of a good day's work — people route around the tool, and it sits unused while the licence renews.

None of these three is anyone's job in the chain. They are exactly the jobs an owner does.

What owning the number looks like

"Owning the number" means one named person is accountable for the business outcome, not for a deliverable along the way.

In practice that person does five things the chain does not.

Agrees the number before anything is bought. Not "deploy a support bot" but "halve the time to first reply on routine tickets without lowering satisfaction". A deliverable can be finished while the number stays flat. A number cannot.
Picks the tool after the workflow is clear. The tool follows the job, not the other way round. Sometimes that means an open-source part, sometimes a paid product, sometimes neither.
Stays for the build. The person who made the plan is in the standups while it is built, so when reality disagrees with the slide, the plan changes that week rather than at the next review.
Watches it after launch. Output checks, spend caps, an owner for every agent, and a regular look at whether the number is moving.
Hands it over properly. Trains the people whose work changed, writes down who decides what, and leaves the business able to run it.

None of these is exotic. They are what a good CTO or head of operations does anyway. The problem for most companies of 20 to 200 people is that they do not have that person, cannot justify one full time, and so they buy the three parts separately and hope the handoffs hold.

One embedded owner instead of three handoffs

The alternative to the chain is not a bigger firm that does all three steps. It is one senior person, inside the team, who owns the plan, the build and the hand-over — and who is still there when the number is measured.

That is the idea behind a Forward Deployed CTO. The role sits inside your team: in standups, in planning, in the decisions. It writes the plan, then builds it with your people, then hands it over. The deck and the build are the same person's work, so there is nothing to lose in the handoff. When the question is the whole business rather than the product — AI across sales, support, finance and operations — the same seat is a Fractional CAIO.

We describe how we run it in three words: Embed. Equip. Hand over. The point of the first word is the one this whole piece is about. Inside the team, not looking in from outside.

Questions to ask before you buy any part of the chain

If you are about to commission an AI plan, buy an AI tool or hire someone to integrate one, these questions are worth asking first. They are useful whoever you end up working with.

What number is this meant to move, and who will be watching it in six months? If the answer is a deliverable, or a team rather than a person, the number is unowned.
Will the people who write the plan be there when it is built? If not, who translates when the build meets reality?
Who owns the tool after launch? Watching output, updating prompts, handling a model change, answering for a bad result.
Who is changing how people work? Training, decision rights, the new process.
What do we own at the end? Code, data, playbooks and the ability to run it without the people who built it. If you would be dependent on them, that is a cost worth pricing now.

A plan with good answers to all five will usually work, whoever writes it. A plan without them is a deck.

Frequently asked questions

Why do AI plans fail in small companies? Usually because nobody owned the result. The plan, the tool and the integration came from three parties, each did its part, and the business outcome belonged to none of them.

What does it mean to own the number? One named person is accountable for the business outcome, agrees it up front, sees it every week and can change the plan when it is not moving.

Is an embedded CTO or CAIO better than a strategy firm? They solve different problems. An outside-in review suits a one-off decision. AI adoption needs someone who stays through the plan, the build and the hand-over.

Working with us

If the chain above sounds familiar, the Forward Deployed CTO page explains how the embedded role works and what it installs. It is a longer read than this one, and it will tell you if the role is not what you need.